devez-vibe@1.9.27
Stable terminal UI for Codex and Claude Agent SDK
AI Security Review
scanned 1mo ago · by lpm-firewall-aiPostinstall unconditionally plants this package's own skills on the broad Codex and Claude user skill surfaces and may pip-install search dependencies. That is first-party agent-extension setup rather than a hidden foreign hijack, because the skills match the published product and the installer shows no credential export or remote code fetch.
Decision evidence
public snapshot- npm postinstall runs install-skills.mjs and prepare-search-deps.mjs with no consent prompt.
- install-skills copies the bundled luna-loop and insane-search trees into the Codex and Claude user skill directories.
- That copy overwrites each skill folder, deletes files the bundle no longer contains, and writes a bundled marker so both agents load these skills.
- When Python 3 is present, postinstall may pip-install curl_cffi and related packages into the current user environment.
- The written skills are this package's own search and review workflows, not a hidden third-party payload.
- The installer does not read credentials or send them to a network endpoint.
- The package does not depend on another release of itself, and the installer is not obfuscated.
Source & flagged code
8 flagged · loading sourcePackage defines install-time lifecycle scripts.
package.jsonView on unpkgInstall-time lifecycle script is not statically allowlisted and needs review.
package.jsonView on unpkgSource creates an unconsented AI-agent control surface through install-time mutation or a default unauthenticated remote skill channel.
Findings
Affected versions and remediation
This report applies to devez-vibe@1.9.27.
See all versions of devez-vibe for other recorded verdicts.
- Review the evidence and your use of devez-vibe@1.9.27 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.